Frequently asked questions

Regulation (EU) 2016/679 (GDPR) clearly distinguishes the two concepts. Article 4(5) defines pseudonymisation as processing personal data so that it cannot be attributed to a person without additional information kept separately. Recital 26 specifies that true anonymisation requires that identification is no longer possible using means reasonably likely to be employed.

For example, in legal submissions, “Ms Dupont” becomes “[PERS1]” and “SAS Acme” becomes “[ORG1]”, with the key stored separately. While the key exists, this is pseudonymisation. If neither a key nor cross-referencing allows re-identification, it may qualify asanonymisation —a threshold rarely reached for a complete litigation matter.

For lawyers, professional secrecy under the CNB's French National Rules of Professional Conduct applies alongside the GDPR: data masking must be controlled before any disclosure to a third party or AI tool. Lexanon performs pseudonymisation (rather than anonymisation): local detection and consistent replacements, with a mapping table for decoding. The data remains personal; users remain responsible for final review and must not present results as GDPR anonymisation.

The CNIL explains that anonymisation aims to make identification practically impossible. The method varies by dataset or document matter, but the steps remain the same.

1. Inventory — first and last names, addresses, phone numbers, emails, IBANs, birth dates, case numbers, organisations and precise locations.
2. Choose the process — deletion, aggregation, generalisation, redaction or pseudonymisation with stable codes.
3. Consistency — same entity, same treatment across all documents to avoid leaks through cross-referencing.
4. Human review — review before disclosure, including headers, footnotes and metadata.
5. Key — if processing is reversible, keep the mapping table separate and secure.

For example, a customer register excerpt contains “Jean Martin, 12 rue de la Paix, 75002 Paris, 06 12 34 56 78”. Strong anonymisation might retain only “individual customer, central Paris, telephone contact”. Pseudonymisation would preserve useful structure: “[PERS1], [ADR1], [TEL1]”, with a separate key.

Be aware: what many call “anonymising a PDF” is often pseudonymisation. Lexanon does not claim anonymisation: it is software for local pseudonymisation (consistent codes plus a mapping table). For AI use, pseudonymise first, query the model with the masked version, then decode the response on your computer.

An administrative document—decision, letter, form, notice or case excerpt—often contains concentrated personal data. Lexanon's intelligent pseudonymisation workflow replaces it with consistent labels such as [PERS1], [ORG1] and [ADR1], retaining a mapping key for later decoding. Everything happens on your computer.

For example, a decision mentions “Mr Karim Benali, born on 14/03/1988 in Lyon, 4 impasse des Lilas” and “SAS Acme”. After Lexanon analysis, the text might show [PERS1], [DATE1], [ADR1] and [ORG1] throughout; the key stays with you. While it exists, this is pseudonymisation, not anonymisation.

Follow this Lexanon workflow when seeking to “anonymise an administrative document” in everyday practice: consistent masking before disclosure. Do not present the result as GDPR anonymisation; the user remains responsible for reviewing before sharing.

Free or open-source data anonymisation software mainly covers manual redaction, spreadsheet scripts or limited masking functions. It often struggles with legal entities, spelling variants, document batches and metadata. For a firm or public body, a single missed page can compromise confidentiality and, for lawyers, professional secrecy as emphasised by the CNB.

For example, manually blacking out “Dupont” in a 120-page PDF leaves “Mr Dupont”, “Mr and Mrs Dupont” and the email j.dupont@… in the file properties untouched. A detection engine reduces this risk without replacing final human review.

Lexanon is neither permanently free software nor a GDPR anonymisation tool: it is software for local pseudonymisation. It offers a 7-day free trial to pseudonymise PDFs, DOCX files, emails and more, with a mapping table and metadata-free export. The subscription then covers updates and support. A free tool can help with occasional non-sensitive files; for regular confidential work, prioritise professional local pseudonymisation without confusing it with anonymisation.

Lawyers and legal professionals want to use generative AI (LLMs) with submissions, contracts, emails and procedural documents without breaching professional secrecy or the GDPR. Pasting a named PDF into ChatGPT, Claude.ai or another cloud assistant exposes names, addresses, IBANs and strategy.Lexanon AI reverses the workflow: local pseudonymisation → send to the model through your API key → local decoding of responses for reading.

For example, you request a summary of “SAS Acme”'s obligations to “Ms Dupont”. The model receives a document containing [ORG1] and [PERS1]. You then read “SAS Acme” and “Ms Dupont” through the mapping table, without sending those names to the AI provider.

1. Pseudonymise documents, individually or as a folder, with intelligent pseudonymisation.
2. Review the preview before attaching documents to the AI discussion.
3. Connect your API key (BYOK): Claude, ChatGPT / OpenAI, Mistral, Gemini, etc.
4. Ask the model about the pseudonymised document or matter (summary, analysis, deliverable).
5. Read the response decoded locally; export or decode a deliverable if needed.

Useful terms: secure AI for lawyers, ChatGPT legal documents, Claude professional secrecy, confidential LLM, AI without data leaks, pseudonymised prompt, anonymised AI (common usage), Bring Your Own Key, local processing before an LLM. Further reading: AI use in Lexanon and Layer (agent interface).

Processing submissions, opposing exhibits, emails and PDF appendices manually, one by one, creates inconsistencies: “Dupont” becomes [PERS1] in one file and [PERS3] in another. Lexanon avoids this through Document Manager and the intelligent pseudonymisation in batch mode: one matter, one key, stable codes.

For example, a 40-document matter repeatedly cites “Paris Commercial Court” and “SAS Acme”. After folder import, Banco and the other categories consistently produce [ORG_TRIB] and [ORG1]. Review once, export the batch, then open Lexanon AI with the same matter context.

Useful terms: pseudonymise a matter, anonymise several documents, batch processing, document consistency, single mapping table, procedural documents, law firm matter, metadata-free export, prepare a matter for AI. Remember: Lexanon performs pseudonymisation, not GDPR anonymisation. Keep the key separate and review before sharing.

Comparing architectures involves more than “local versus SaaS”. An anonymisation or pseudonymisation tool reads names, organisations, IBANs, addresses, dates and identifiers in depth: precisely the sensitive core of the matter. Sending it to a third-party server, even a “secure” one, expands the attack surface, processor chain and risks of foreign legal access (CLOUD Act, FISA, etc.) if the provider or host is subject to US law.

Cloud offerings need not be rejected in principle. They may be considered only if the provider transparently documents where engines run, which processors handle content, whether documents are retained, for how long and in which country, and whether every part of the chain—publisher, cloud, support and telemetry—is outside the US CLOUD Act. Without that clarity, assume the risk is too high for litigation, professional secrecy or sensitive personal data.

For example, uploading submissions to an “AI anonymisation” SaaS hosted by a US hyperscaler entrusts an opaque chain with automatically extracting everything you wanted to protect. With a local architecture such as Lexanon, analysis stays on the computer; no document is sent away for pseudonymisation.

Cloud is not prohibited, but the burden of proof lies with the provider: tools, hosting and absence of CLOUD Act exposure. For professional secrecy and sensitive matters, local remains the preferred architecture. See also Lexanon local software and secure AI with pseudonymised documents.

Useful terms: local versus cloud pseudonymisation, SaaS anonymisation, CLOUD Act, digital sovereignty, France / EU hosting, GDPR processor, data leak, lawyers' professional secrecy, offline processing, on-premises, zero upload.